Papers taken apart step by step, AI news in brief, and news from Pixelia.
The papers and models that change how AI goes to production, explained step by step.
Open the section →AI news in brief: releases, numbers and announcements that matter for production.
Open the section →Products, partnerships and behind the scenes of our projects.
A Stanford paper is going around with one number: 31 times faster. The posts relaying it announce the end of language models as we know them. The paper is serious and the idea is elegant. But the number does not compare what people think, and the two objects are not in the same league.
Read the article →TypeSafe AI released Jev on September 15. It cannot write a sentence, code, or reason step by step. It does one thing: pick among options you give it, and say how sure it is. That is exactly what large models were missing before anyone would let them decide on their own.
Read the article →DeepSeek just released V4.1 Flash, and it is a feat: an open model that holds its own against the best models of the moment, built by a team forty times smaller than OpenAI's, with no access to the latest Nvidia chips or giant datacenters. So how did they do it?
Read the article →
